My solution when it failed installing

[goodevile]

I am tired to explain everything so here it's what I did:

First install CUDA 12.5 (cuda cuda cuda cuda 🎶)

https://developer.nvidia.com/cuda-downloads

Then:

git clone git@github.com:facebookresearch/segment-anything-2.git

cd segment-anything-2

Next: pip install wheel setuptools pip install ninja

why ninja? well... got this error: UserWarning: Attempted to use ninja as the BuildExtension backend but we could not find ninja.. (don't ask me, ask who wrote this code that asks for ninjas 🥷)

Are you using a venv? (virtual environment) then check nvcc version:

nvcc --version

12.* ? then you are good to go and use pip install -e .

not 12.*? then check where cuda is installed:

ls /usr/local/

are you seeing the cuda folder like this?:

bin cuda cuda-12 cuda-12.5 etc games include lib man sbin share src

then do this:

export PATH=/usr/local/cuda-12.5/bin${PATH:+:${PATH}} export LD_LIBRARY_PATH=/usr/local/cuda-12.5/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

deactivate your environment and activate it again. then:

pip install -e .

You happy? leave a like. Unhappy? I tried ok?

[TimoK93]

Thanks, that helped! I wanted to add that I needed to set the GCC version to something between 9 < Version <14. You can do this by

CC=gcc-11 pip install -e .

or with an other available version that exists on your device!

[seanwhen]

Which version of python is the conda environment you are using? Is it 3.9 or 3.10?

[TimoK93]

I use conda with python 3.10.

[fritol]

worked for me:

install torch nightly! install CUDA 12.1 run the sam2 pip command

[mahmoodkashmiri]

Try this